Ansari Saleh Ahmar
This study presents a comprehensive comparative analysis of five distinct time series forecasting methodologies—HySutte (HybridSutte), TBATS (Trigonometric Box-Cox ARMA Trend Seasonal), ETS (Exponential Smoothing State Space Model), ARIMA (Autoregressive Integrated Moving Average), and Random Forest—for predicting Indonesia’s monthly import values. Using monthly import data from January 2021 to December 2024, with 40 months for training and 8 months for testing, we evaluate model performance through RMSE, MAE, and MAPE metrics. The HySutte model incorporates hybrid approaches combining selective outlier handling using Median Absolute Deviation, trend-residual decomposition, enhanced α-Sutte forecasting, pattern matching algorithms, and adaptive weight integration. Results demonstrate that HySutte significantly outperforms all benchmark models, achieving RMSE of 1392.31, MAE of 1037.75, and MAPE of 4.97% on test data, representing improvements of 28.1% over TBATS, 47.3% over ETS, 37.8% over ARIMA, and 28.1% over Random Forest in RMSE. Statistical significance testing using the Diebold-Mariano test confirms that HySutte's superior performance is statistically significant at the 5% level against all benchmarks. The study reveals HySutte’s superior robustness to outliers and its ability to capture both short-term fluctuations and long-term patterns in volatile economic time series. Economic impact analysis shows HySutte generates the lowest total economic cost of 281.9 million USD through better balance between inventory holding costs and stockout penalties. These findings provide valuable insights for policymakers and practitioners in international trade forecasting and inventory management. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2026.
Department of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Negeri Makassar, Makassar, 90223, Indonesia
